원심펌프의 고장 상태 진단을 위한 머신러닝 알고리즘 연구 = A Study on Machine Learning Algorithms for Diagnosing Fault Conditions in Centrifugal Pumps

다국어 초록 (Multilingual Abstract)

Centrifugal pumps are used widely to deliver various fluids, such as water, oil, and chemicals, and are positioned as key equipment in industrial sites. Nevertheless, various failures that can prevent the pump from operating normally, such as seal cracks and bearing wear, can occur if a centrifugal pump is operated for a long time, is overloaded, or there is no pre-maintenance, resulting in large losses and maintenance costs. Diagnosing and preventing these failures in advance is essential. Therefore, this study confirmed the main failure status of the centrifugal pump based on its maintenance history and FMEA analysis. Data on normal, bearing wear, and seal crack status were collected, and the dimensions of the data were reduced by applying the PCA, TSNE, UMAP, and LLE dimension-reduction techniques. In addition, the SVM, KNN, DT, and XGBoost classification algorithms were applied based on the reduced data, and the performance was compared based on the performance indicators, Precision, Recall, F1-score, and Accuracy using the evaluation data. Among these performance indicators, the algorithm with the most appropriate classification performance between states was selected.